Wim Mulder (Shell. Netherlands)

Time- versus frequency-domain modelling of seismic wave propagation

Abstract

Realistic modelling of elastic wave propagation is feasible on present-day parallel computers when using time-domain finite-difference methods. Within Shell, a parallel code for acoustic and elastic modelling has been in use seen 1993. It is routinely applied to two-dimensional problems but still rarely to 3D problems because of its computational cost.

Frequency-domain methods are competitive for two-dimensional problems, although the gain in speed is not sufficiently dramatic to justify the increase in code complexity. In three dimensions, iterative solvers appear to be necessary. An number of experiments with BICGSTAB2 and a preconditioner based on separation of variables show fast convergence for horizontally layered models but have an unacceptable performance degradation in more complex models for higher frequencies. Still, frequency domain methods may be attractive for migration and inversion, if only a reduced number of frequencies are used.